Wednesday, April 2, 2014

Build a Smarter Group from Scratch: Converse Equitably, Add Women, Stir.

The below is a guest post from Dr. Sarah Ballard. Dr. Ballard completed her PhD in Astronomy & Astrophysics at Harvard University in 2012 and is now a NASA Sagan fellow at the University of Washington.  Follow her on twitter at: @hubbahubble

To work in astronomy is now to work in teams. A recent PNAS study reported that the average team size associated with a single publication grew from 1.5 in 1961-1965, to 6.7 in 2006-2010 (Milojević 2014). However, much of the dialog about the nature of intelligence is still focused upon single individuals. Though the conception of a person’s intelligence as a fixed quantity is fraught at best (see this summary by J. Johnson), it’s often the only way we conceive of intelligence at all. What is the nature of the intelligence of a group? What quantities are predictive of it, if any? It is now groups of individuals who publish new ideas in our field. To ask about group intelligence is now to ask: “How are units of knowledge produced?”

A study in Science by Woolley et al. (2010) investigated this question. They define group intelligence in the following way. A metric for group intelligence should (a) reflect “how well a single group can perform a wide range of different tasks”, and (b) “predict how that same group will perform other tasks in the future” (Woolley et al. 2010). We measure the intelligence of individuals with similar assumptions. As you read this article on the Women in Astronomy blog, please pause here. What are the quantities you predict will reflect group intelligence? Is it the intelligence of the “smartest” individual, who might direct the group well? Is the mean intelligence of the individuals? Perhaps the intelligence of the “least intelligent” individual is correlated with the intelligence of a group. I will posit here, as my own hypothesis, that most astronomers assume that one of the first two metrics will be predictive.

The authors first created random groups of three individuals. These people worked together for up to five hours on a set of simple group tasks, and one complex task. The nature of these tasks included “puzzles, brainstorming, making collective moral judgments, and negotiating over limited resources” (Woolley et al. 2010). I trust the reader will agree with me when I say that real astronomers in groups grapple regularly with tasks of this kind. The second study involved groups of 2-5 individuals, to test whether any group intelligence parameter is robust to small changes in group size.

They found that no metric related to individual cognitive ability is meaningfully correlated with the intelligence of a group. However, there does exist strong evidence for a single dominant factor underlying group intelligence, which that authors denote by the variable c. If not derived from the intelligence of members in the group, what is this factor? The factors that the authors, and probably many of us, might guess: group cohesion, motivation, or satisfaction, were also not predictive. Rather, the authors identified three variables that were significantly correlated with c.
  1. The average social sensitivity of group members. This was measured, as described in a published study by Baron-Cohen et al., by the “Reading the Eyes in the Mind” test. The subject is presented with photographs of the eye region of different individual people, and “is asked to choose which of two words best describes what the person in the photograph is thinking or feeling.” Correctly identifying the emotional state evinced by the person in the photograph is predictive of “social intelligence.” Baron-Cohen et al. (2001) note that this psychiatric metric overlaps with the term “empathy.” 
  2. Equal distribution of conversational turn-taking. Groups with a single individual dominating the conversation were less collectively intelligent.
  3. Proportion of women in the group. Fractionally more women in the group was “positively and significantly” correlated with group intelligence. The authors note that this parameter is almost entirely degenerate with the “social sensitivity” parameter described above. Consistent with previous research, women generally perform better on this metric than their male counterparts. 
The authors go on to conclude that group intelligence is “easier to raise” than the intelligence of an individual person. It can result from mild social engineering alone.

I close by observing that the groups of astronomers I have most often encountered do not evince (2) or (3) with any regularity. I leave it to the reader to conjecture whether individual members of the groups they’ve observed exhibit the “social sensitivity” described by factor (1). In this sense, our basic unit of knowledge creation in astronomy, the small group, is functionally sub-optimally.

Further Reading